Geração procedural de melodias por gênero usando aprendizado de máquina

Abstract

The procedural generation of melodies can be understood as the processes through which the composition of musical notes sequences can be automated. For computer science, especially in the field of machine learning, this goal can be achieved through statistical sampling since the set of keys of contemporary musical systems and the relationships between notes are organized due to both their physical and mathematical properties. However, due to the strictly computational nature of most of the implementations described in the current literature, the approval rating of such melodies both by specialists and the general public is still very low. Thus, the present work implements a web platform that contemplates the procedural generation of melodies by genre using Neural Networks (due to their capacity to identify trends, which makes them able to understand and apply rules of multiple musical systems), and the rating of the generated melodies. Considering the three genres available on the platform, the system managed to reach a positive rating in almost a third of the melodies generated.